Tri Arief Sardjono

Automatic Cobb angle determination from radiographic images.

Study Design. Automatic measurement of Cobb angle in patients with scoliosis. Objective. To test the accuracy of an automatic Cobb angle determination method from frontal radiographical images. Summary of Background Data. Thirty-six frontal radiographical images of patients with scoliosis. Methods. A modified charged particle model is used to determine the curvature on radiographical spinal images. Three curve fitting methods, piece-wise linear, splines, and polynomials, each with 3 variants were used and evaluated for the best fit. The Cobb angle was calculated out of these curve fit lines and compared with a manually determined Cobb angle. The best-automated method is determined on the basis of the lowest mean absolute error and standard deviation, and the highest R2. Results. The error of the manual Cobb angle determination among the 3 observers, determined as the mean of the standard deviations of all sets of measurements, was 3.37°. For the automatic method, the best piece-wise linear method is the 3-segments method. The best spline method is the 10-steps method. The best polynomial method is poly 6. Overall, the best automatic methods are the piece-wise linear method using 3 segments and the polynomial method using poly 6, with a mean absolute error of 4,26° and 3,91° a standard deviation of 3,44° and 3,60°, and a R2 of 0.9124 and 0.9175. The standard measurement error is significantly lower than the upper bound found in the literature (11.8°). Conclusion. The automatic Cobb angle method seemed to be better than the manual methods described in the literature. The piece-wise linear method using 3 segments and the polynomial method using poly 6 yield the 2 best results because the mean absolute error, standard deviation, and R2 are the best of all methods. Level of Evidence: 3

A framework for human spine imaging using a freehand 3D ultrasound system

The use of 3D ultrasound imaging to follow the progression of scoliosis, i.e., a 3D deformation of the spine, is described. Unlike other current examination modalities, in particular based on X-ray, its non-detrimental effect enables it to be used frequently to follow the progression of scoliosis which sometimes may develop rapidly. Furthermore, 3D ultrasound imaging provides information in 3D directly in contrast to projection methods. This paper describes a feasibility study of an ultrasound system to provide a 3D image of the human spine, and presents a framework of procedures to perform this task. The framework consist of an ultrasound image acquisition procedure to image a large part of the human spine by means of a freehand 3D ultrasound system and a volume reconstruction procedure which was performed in four stages: bin-filling, hole-filling, volume segment alignment, and volume segment compounding. The overall results of the procedures in this framework show that imaging of the human spine using ultrasound is feasible. Vertebral parts such as the transverse processes, laminae, superior articular processes, and spinous process of the vertebrae appear as clouds of voxels having intensities higher than the surrounding voxels. In sagittal slices, a string of transverse processes appears representing the curvature of the spine. In the bin-filling stage the estimated mean absolute noise level of a single measurement of a single voxel was determined. Our comparative study for the hole-filling methods based on rank sum statistics proved that the pixel nearest neighbour (PNN) method with variable radius and with the proposed olympic operation is the best method. Its mean absolute grey value error was less in magnitude than the noise level of a single measurement.

Braille character recognition using find contour method

Braille letters is characters designed for the blind, composed of six embossed points, arranged in a standard braille character. Braille letters is touched and read using fingers, therefore the sensitivity of the fingers is important. Those characters need to be memorized, so it is very difficult to be learned. The aim of this research is to create a braille characters recognition system and translate it to alpha-numeric text. Webcam camera is used to capture braille image from braille characters on the paper sheet. Cropping, grayscale, thresholding, erotion, and dilation techniques are used for image preprocessing. Then, find contour method and image recognition lookup table method are used to recognize the braille characters. The system can recognize braille characters with 100% accuracy even when the braille image is tilted up to 0.5 degrees.

Visual ball tracking and prediction with unique segmented area on soccer robot

Object detection and tracking system has been developed by several researchers. This paper present algorithm for visual ball detection and ball estimation for goalie (goalkeeper) robot. The ball is captured by a camera with a fish-eye lens and processed for detection and tracking. Images from fish-eye camera are curved images. Images are thresholded to Hue Saturation Value (HSV). The system can predict goal area and ball position with multilayer backpropagation neural network (BPNN). The BPNN inputs are x and y axis of the ball. The BPNN outputs are goal area prediction and ball area prediction. The training data is unique segmented area. According to the changes of previous ball distance, the system will predict the direction of the next ball position. The achievement result (unique kernel 3×3, MSE <0.001, 30 samples data) for ball position prediction is 76.67%. The achievement result (unique kernel 3×3, MSE <0.001, 30 samples data) for goal area prediction is 100%.

Pattern Matching Performance Comparisons as Big Data Analysis Recommendations for Hepatitis C Virus (HCV) Sequence DNA

A data bank can provide very useful information while mined properly.[27] In order to be optimally extracted, data mining can be done by observing capacity and characteristics of the data; so it can generates Knowledge Discovery in Databases as expected. For instance in Gene Bank, every single record of DNA, there are at least ten thousand sequences recorded. If the data is more than a hundred records, it will be a big sequence of data to be processed. Hepatitis C Virus (HCV) is a liver disease which can infect humans through blood. HCV infection can be asymptomatic, or it can be hepatitis acute, chronic, furthermore cirrhosis. Hepatitis C is generally does not show symptoms in the early stages. About 75 percent people with hepatitis C did not realize that they had infected until liver damage years later. Therefore needed a sequences DNA Mining is needed to analyse the DNA history whether it is infected by HCV or not. This study compares several methods of string matching to discover which methods have the best performance in processing DNA mining. In addition, this study also analyzed DNA HCV genetic mutations trend as a Knowledege Discovery in Database in DNA mining.

Identification of phonocardiogram signal based on STFT and Marquart Lavenberg Backpropagation

Auscultation is one of method that has been used by doctor in the process to identify heart disease. But with only using auscultation is not possible to obtain quantitative data, because relating to the skills, experience and subjectivity doctor, for it to be made a program that can help doctors to identify heart abnormalities. Short Time Fourier Transform is used to analyze the frequency and timing of the patterns of normal and abnormal heart sounds. With the method of Levenberg Marquardt Back propagation expected 10 class heart sound can be identified with a good. The results showed that the accuracy will be higher when the number of neurons increased. When neurons 25 produce more accurate results with accuracy percentage is higher and more stable for each class. Where the highest accuracy is on the holo diastolic murmur class and the Normal Heart class with accuracy up to 99,9999%.

A signal processing framework for multimodal cardiac analysis

The heart is a complex organ in the cardiovascular system which its measurement and analysis system in clinical level should be realized in an integrated system including all cardiac vital signs. A previous study combined ECG and PCG analysis could detect murmur symptom. However, the heart mechanical activity could not be described. This study developed a multimodal analysis of cardiac signals consisting of ECG signals, carotid pulse, and PCG. The purpose of this study was to develop and test an appropriate signal processing framework to facilitate parameter extraction and to enhance understanding of underlying mechanisms in the cardiac physiology. Frequency and time-frequency domain analysis of cardiac signals were performed to design sophisticated digital filters. Recursive digital filters were chosen in realizing segmentation methods and the advanced signal processing techniques were performed in parameter extraction. Results show the proposed method was able to detect QRS complex, P and T waves in ECG signal with 88% sensitivity and also percussion wave with 85.62% sensitivity. Sistolic (S1) and diastolic (S2) heart sound also could be separated. Classification of normal and the disease type of heart based on the cardiac parameters resulted by the presented signal processing framework would be next research topic.

Six key points lip's feature extraction using adaptive threshold segmentation

Key points of lip give important cues of lip shape. The application of lips shape tracking could be used for speech recognition, lip reading and many multimedia applications. In this paper, a novel adaptive threshold segmentation for six key points lips feature extraction algorithm is proposed. First, Viola Jones method is used to detects face in an image frame, after that the ROI of mouth is set. Color transformation in RGB space and adaptive threshold are used for lip segmentation. Further step, contour of the segmented lip is defined and filled in with certain color. Finally, the six key points, which are left and right corners, lower point, and three points of the Cupidons bow of lip are scanned. The performance of the proposed method is satisfactory and compared with existing methods it brings a significant improvement in accuracy.

Pengenalan Viseme Dinamis Bahasa Indonesia Menggunakan Convolutional Neural Network

There has been very little researches on automatic lip reading in Indonesian language, especially the ones based on dynamic visemes. To improve the accuracy of a recognition process, for certain problems, choosing suitable classifiers or combining of some methods may be required. This study aims to classify five dynamic visemes of Indonesian language using a CNN (Convolutional Neural Network) and to compare the results with an MLP (Multi Layer Perceptron). Varying some parameters theoretically improving the recognition accuracy was attempted to obtain the best result. The data includes videos on pronunciation of daily words in Indonesian language by 28 subjects recorded in frontal view. The best recognition result gives 96.44% of validation accuracy using the CNN classifier with three convolution layers.

Optimasi Support Vector Machine untuk Memprediksi Adanya Mutasi pada DNA Hepatitis C Virus

Hepatitis C Virus (HCV) is a virus which capable of infecting RNA that can lead to changes in the DNA sequence. This change of DNA arrangement is called genetic mutation. Every mutation occurs in HCV, it will be called a new subtype. Over time, HCV subtypes increase, and will continue to grow as the HCV mutation cycle progresses faster. Therefore, a way to find a mutation in millions of sequences in the gene bank is needed. This study tested six types of Support Vector Machine (SVM) methods to determine the best SVM kernel performance in the application of HCV DNA sequence detection in isolated. The tested SVM kernel was linear, quadratic, cubic, fine Gaussian, median Gaussian, and coarse Gaussian. The data set is 1000 isolated DNA consisting of 500 isolated Homo Sapiens and 500 isolated HCV. Firstly, the data set will go through the pattern search process using the Edit Levenshtein Distance method, then the result of the processing will be the variable x in SVM. The target or variable y on SVM is the positive or negative value of the isolated against HCV. The results showed that among the six types of SVM methods being tested, the method of fine Gaussian SVM has the lowest performance of 77.4%. The SVM method was tested by performing optimizations on the determination of the hyperplane. The test results proved that the SVM method was able to analyze the presence of HCV mutations in isolated DNA with an accuracy of 99.8%.